Doug can you please elaborate on this. I thought each segment maintained its own list of deleted documents (since segments are WRITE ONCE, and when that segment is merged or optimized it would "go away" anyway, as the deleted documents are removed.
In my reopen() implementation, I check to see if the existing segment name is the same as an already open segment, and then just use the existing SegmentInfo object (since it should still have reference to its deleted documents). For example, Index has 3 (1-3) segments. A new document is written that causes a segment to be created (4). A reopen would retain the SegmentInfo for 1-3, and create a new one for 4. It would be no different if segment 2 had deleted documents when the creation of segment 4 occurs, segment 2 is not modified in this case. If adding the new document, which creates a new segment, caused a merge, segment 2 would be rewritten (and the deletions processed), so the segment name for 2 would no longer be valid anyway, and the SegmentInfo would not reused. I've had this code in production for almost 2 years and have not seen any problems - trying to get a handle on the possibility that our code may be "unstable". -----Original Message----- From: Doug Cutting [mailto:[EMAIL PROTECTED] Sent: Wednesday, April 26, 2006 3:44 PM To: [email protected] Subject: Re: GData, updateable IndexSearcher jason rutherglen wrote: > I was thinking you implied that you knew of someone who had customized their own, but it was a closed source solution. And if so then you would know how that project faired. I don't recall the details, but I know folks have discussed this previously, and probably even posted patches, but I don't think any of the patches was ready to commit. > Wouldn't there also need to be a hack on the IndexWriter to keep track of new segments? I think the 'public static IndexReader.reopen(IndexReader old)' method I proposed can easily compare the current list of segments for the directory of old to those that old already has open, and determine which can be reused and which new segments must be opened. Deletions would be a little tricky to track. If a segment has had deletions, then a new SegmentReader could be cloned from the old, sharing everything but the deletions, which could be re-read from disk. This would invalidate cached filters for segments that had deletions. You could even try to figure out what documents have been deleted, then update filters incrementally. That would be fastest, but more complicated. Doug > ----- Original Message ---- > From: Doug Cutting <[EMAIL PROTECTED]> > To: [email protected] > Sent: Wednesday, April 26, 2006 11:27:44 AM > Subject: Re: GData, updateable IndexSearcher > > jason rutherglen wrote: > >>Interesting, does this mean there is a plan for incrementally updateable IndexSearchers to become part of Lucene? > > > In general, there is no plan for Lucene. If someone implements a > generally useful, efficient, feature in a back-compatible, easy to use, > manner, and submits it as a patch, then it becomes a part of Lucene. > That's the way Lucene changes. Since we don't pay anyone, we can't make > plans and assign tasks. So if you're particularly interested in this > feature, you might search the archives to find past efforts, or simply > try to implement it yourself. > > I think a good approach would be to create a new IndexSearcher instance > based on an existing one, that shares IndexReaders. Similarly, one > should be able to create a new IndexReader based on an existing one. > This would be a MultiReader that shares many of the same SegmentReaders. > > Things get a little tricky after this. > > Lucene caches filters based on the IndexReader. So filters would need > to be re-created. Ideally these could be incrementally re-created, but > that might be difficult. What might be simpler would be to use a > MultiSearcher constructed with an IndexSearcher per SegmentReader, > avoiding the use of MultiReader. Then the caches would still work. > This would require making a few things public that are not at present. > Perhaps adding a 'MultiReader.getSubReaders()' method, combined with an > 'static IndexReader.reopen(IndexReader)' method. The latter would > return a new MultiReader that shared SegmentReaders with the old > version. Then one could use getSubReaders() on the new multi reader to > extract the current set to use when constructing a MultiSearcher. > > Another tricky bit is figuring out when to close readers. > > Does this make sense? This discussion should probably move to the > lucene-dev list. > > >>Are there any negatives to updateable IndexSearchers? > > > Not if implemented well! > > Doug > > > > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
